笔记-AlexNet

AlexNet共分为5个卷积层和3个全连接层

1、conv1(卷积层）

将96像素层分为两组，每组在独立GPU上运行

2、conv2(卷积层）

第二层输入数据为第一层输出的27*27*96的像素层，为便于后续处理，每幅像素层的左右两边和上下两边都要填充2个像素；

3、conv3(卷积层)

为便于后续处理，每幅像素层的左右两边和上下两边都要填充1个像素；

4、conv4(卷积层)

5、conv5(卷积层）

6、全连接层1

7、全连接层2

8、全连接层3

基于keras实现代码如下：

    model = Sequential()

    ## Architecture

    model.add(
        Conv2D(filters=96, kernel_size=(11, 11), strides=4, padding='same', activation='relu', input_shape=(64, 64, 1),
               kernel_initializer='he_normal'))
    model.add(MaxPooling2D(pool_size=(3, 3), strides=(2, 2), padding='same', data_format=None))  # overlapping pooling

    model.add(Conv2D(filters=256, kernel_size=(5, 5), strides=1, padding='same', activation='relu',
                     kernel_initializer='he_normal'))
    model.add(MaxPooling2D(pool_size=(3, 3), strides=(2, 2), padding='same', data_format=None))

    model.add(Conv2D(filters=384, kernel_size=(3, 3), strides=1, padding='same', activation='relu',
                     kernel_initializer='he_normal'))
    model.add(Conv2D(filters=384, kernel_size=(3, 3), strides=1, padding='same', activation='relu',
                     kernel_initializer='he_normal'))
    model.add(Conv2D(filters=256, kernel_size=(3, 3), strides=1, padding='same', activation='relu',
                     kernel_initializer='he_normal'))
    model.add(MaxPooling2D(pool_size=(3, 3), strides=(2, 2), padding='same', data_format=None))

    model.add(Flatten())
    model.add(Dense(units=4096, activation='relu'))
    model.add(Dense(units=4096, activation='relu'))

    model.add(Dense(units=1000, activation='softmax'))
    model.summary()

    model.compile(loss=keras.losses.categorical_crossentropy,
                  optimizer=keras.optimizers.Adadelta(),
                  metrics=['accuracy'])

    model.fit(image_train, label_train,
              batch_size=batch_size,
              epochs=epochs,
              verbose=1,
              validation_data=(image_test, label_test))